Actin, one of the most abundant proteins in the eukaryotic cell, also has an abundance of relatives in the eukaryotic proteome. To date though, only five families of actins have been characterized in bacteria. We have conducted a phylogenetic search and uncovered more than 35 highly divergent families of actin-like proteins (Alps) in bacteria. Their genes are found primarily on phage genomes, on plasmids and on integrating conjugative elements, and are likely to be involved in a variety of functions. We characterize three Alps and find that all form filaments in the cell. The filaments of Alp7A, a plasmid partitioning protein and one of the most divergent of the Alps, display dynamic instability and also treadmill. Alp7A requires other elements from the plasmid to assemble into dynamic polymers in the cell. Our findings suggest that most if not all of the Alps are indeed actin relatives, and that actin is very well represented in bacteria.
Actin is present in all eukaryotic cells and is the most abundant protein of the eukaryotic cytoskeleton. Actin participates in such fundamental processes as cell motility, endocytosis, cell remodelling, cytokinesis and transcription (Le Clainche and Carlier, 2008; Pollard and Borisy, 2003; Girao et al., 2008; Wanner and Miller, 2007; Pollard, 2008; Chen and Shen, 2007). Actin is extremely well conserved. The cytoskeletal actins of chicken, cow and man are identical to each other across all 375 amino acids of the protein. The actin of Saccharomyces cerevisiae is exactly the same length, and its sequence is 89% identical to this vertebrate sequence. This level of sequence conservation is not required however, for the actin fold. Actin is a member of a large superfamily of proteins that share the same fundamental architecture. In this superfamily are the 70 kDa heat shock proteins and a group of sugar and sugar alcohol kinases that includes hexokinase and glycerol kinase (Kabsch and Holmes, 1995; Flaherty et al., 1991; Bork et al., 1992). The actin folds of rabbit skeletal muscle actin and the 70 kDa heat shock protein from cow, two members of this superfamily, are only 16% identical at the amino acid sequence level, but can be superimposed with a root mean square deviation of 2.3 Å (Flaherty et al., 1991).
Long assumed to lack a cytoskeleton or cytoskeletal proteins, bacteria have in the last decade been shown to contain homologues of actin and also of tubulin and intermediate filaments (Pogliano, 2008; Graumann, 2007). To date five distinct families of actin-like proteins have been identified in bacteria, and they are no more related to each other than they are to actin (∼ 13% sequence identity). The crystal structures of members of three of these families, of FtsA, MreB and ParM, confirmed that their classification as members of the actin family was appropriate despite the very slight resemblance of their sequences to that of actin (van den Ent and Löwe, 2000; van den Ent et al., 2001; 2002).
With a mere five families of distant relatives identified, actin would appear to have only very sparse representation in bacteria. There are in contrast a great number of actin relatives that have been identified in eukaryotes, and even among these eukaryotic proteins there is considerable sequence and functional diversity. The actin-related proteins, or Arps, were discovered about 20 years ago. Although there exist structures for only Arp2 and Arp3, the secondary structural elements of the actin fold appear to be present in all of the Arps (Muller et al., 2005). Arp1, a component of the dynein activator complex, is the closest to actin in amino acid sequence; the sequences of S. cerevisiae Arp1 and actin are 46% identical. Arp1 retains the signature property of actin: Arp1 polymerizes into filaments with the pitch of filamentous actin. Arp1 also binds ATP, and filament formation, as in actin, is accompanied by ATP hydrolysis. But there are important differences too. Kinetic profiles indicate that there is no barrier to nucleation and that the Arp1 filaments cannot be extended beyond a specific length (Bingham and Schroer, 1999). The divergence is greater for Arp2 and Arp3, which in Saccharomyces are, respectively, 39% and 32% identical to actin. Their crystal structures, which were solved in the context of the bovine Arp2/3 complex, revealed that the actin fold is well preserved in both proteins (Robinson et al., 2001; Nolen et al., 2004). But neither protein homopolymerizes into filaments, each binds ATP with three orders of magnitude lower affinity than actin does, and Arp3 does not appear to hydrolyse ATP at all (Dayel et al., 2001; Dayel and Mullins, 2004). The remaining Arps diverge still further from actin. The sequences of Saccharomyces Arp9 and actin, for example, share only 14% identity, on the order of the bacterial actins.
A recent survey of a single eukaryotic genome, Dictyostelium discoideum, turned up in addition to 17 copies of the actin gene, 16 other genes that code for proteins that closely resemble actin, as well as eight Arps (Joseph et al., 2008). Yet in the entire bacterial kingdom, only five families of actin relatives have been characterized. Here we use a bioinformatics approach and identify more than 35 families of actin relatives in bacteria, most of which have gone unrecognized as actins to date. We characterize one protein from each of three families, and we find that all three proteins form filaments in vivo. We investigate one of these proteins in detail, and we find that it forms filaments with dynamic properties, and that this process requires other elements from the plasmid on which it is encoded. That all three proteins that we sampled give rise to filaments suggests that most if not all of the proteins we found are indeed bacterial actins and that the bacterial ‘actinome’ is therefore much more extensive than previously appreciated.
Identification of more than 35 new families of bacterial actin
Five families of actins have been characterized in bacteria. Our recent work, which led to the discovery of one of these families (Becker et al., 2006), also led us to wonder whether bacteria contain still other uncharacterized actin families. A bioinformatics approach was used to address this question. A blast search was conducted with our recently discovered fifth family member AlfA. Potential new actin sequences that were identified and that were distinct from the five families but still more closely related to actin than to Hsp70 or to the sugar kinases were then used to begin a second round of blast searches. New sequences from the second round of searches were used for a third round, and the searches were continued in this manner for several more rounds. A phylogenetic tree that was generated from these new sequences and the five already identified bacterial actin families revealed that the new sequences comprised more than 35 distinct families of bacterial actins that were only distantly related to each other, to MreB, FtsA, ParM, AlfA and MamK, and to actin itself (Fig. 1A and Table S1). Although each family shares less that 30% identity with the other families, in each sequence could be found the five actin signature motifs of amino acids that are involved in the binding and hydrolysis of ATP (Bork et al., 1992). We have therefore designated these proteins ‘actin-like proteins’ or ‘Alps’ (Fig. 1A).
A remarkable feature of these Alp families is their phylogenetic distance from one another. A single blast search with one of these proteins falls far short of revealing the expanse of the tree, turning up members of only a few of the other Alp families. A blast search with any member of the Alp7 family, for example, fails to identify the established bacterial actins such as MreB or ParM as statistically significant relatives, and a pairwise alignment between the Alp7 family member Alp7A (see below) and either MreB or ParM explains this failure. Alp7A is only 13% identical to MreB and to ParM; it is 11% identical to the entirely unrelated LacI, a protein of about the same length. Nevertheless, the Alp7 family members and all of the other proteins of the tree contain the five conserved motifs of the actin nucleotide binding pocket (Bork et al., 1992), and they could be linked phylogenetically to MreB and to ParM if not immediately, than through intermediates in the form of members of other Alp families. The proteins of the tree are all of roughly the same length, about 350 amino acids, and none of them appear to be more closely related either to Hsp70 or to hexokinase.
The annotations accompanying the sequences indicated that the functions of many of these proteins were unknown. Although a few of the genes appeared to be on bacterial chromosomes, for example, the members of the Alp32 family, most were on phage genomes, plasmids and integrating conjugative elements (Table S1). Given the great phylogenetic divergence among the Alps, it remained possible that these proteins shared nothing more than the ability to bind nucleotide in the manner of actin. We therefore sought to determine whether the Alps were truly actins by looking at their polymerization properties within the cell. We chose three Alp sequences, each from a distinct family of our phylogenetic tree. We fused gfp to the respective genes, and we examined the resulting fusion proteins in E. coli. These genes were gp207 of Bacillus thuringiensis phage 0305φ8-36, from the Alp6 family (Thomas et al., 2007); OrfB from B. subtilis natto plasmid pLS20, from the Alp7 family (Meijer et al., 1995); and orf250 of Proteus vulgaris plasmid Rts1, from the Alp8 family (Murata et al., 2002). As was typical of representatives of these divergent Alp families, these proteins, which we have for simplicity designated Alp6A, Alp7A and Alp8A, shared less than 22% amino acid sequence identity with one another (average of 17.6 ± 3.4%), but actin signature motifs could be found in all three (Fig. 1B). When produced in E. coli, each protein assembled into long filamentous structures that in many cases extended longitudinally through several cells and caused them to grow abnormally as chains in culture (Fig. 1C, D and F). The Alp7 family representative, pLS20 OrfB (Alp7A), was also produced without a GFP tag, and gave rise to chained cells as well (Fig. 1E).
Alp7A is required for plasmid stability
Even though their sequences share only a tenuous resemblance to that of actin, these three proteins, in the absence of any other elements from the source DNA or from the native host, assembled into filamentous structures in E. coli. Like actin, they could polymerize, and they could do so without auxiliary factors when produced at what we assume to be greater than their normal physiological concentrations (Tobacman and Korn, 1983). In order to illuminate the connection between these proteins, their polymerization properties, their function and actin, we chose to study one in detail. The functions of all three proteins were unknown, but the Alp7 family member Alp7A appeared to be a plasmid stability determinant. Actin-like proteins such as ParM are the nucleotide-binding components of one of the two major sets of bacterial plasmid partitioning systems. The genetic organization of these systems is typically tripartite, with a gene that codes for an ATPase, a gene that codes for a DNA-binding protein, and a centromere-like site (Gerdes et al., 2004). This organization is recapitulated here (Fig. 2A). The gene for Alp7A appears to be co-transcribed with a downstream gene, alp7R, that codes for a 134 amino acid protein whose small size and high percentage of charged residues recall the DNA-binding protein ParR (Fig. S1). The putative alp7AR operon is situated near the origin of replication, as is frequently the case for plasmid partitioning systems.
We constructed a plasmid to test for a role of Alp7A in plasmid partitioning. As previous work had shown that the pLS20 origin region is sufficient for replication (Meijer et al., 1995), our plasmid contained both the pLS20 origin of replication and the alp7AR operon (Fig. 2A). The steady state level of Alp7A in a strain containing this mini-pLS20 plasmid matched that in a strain containing pLS20 itself, demonstrating that Alp7A expression is equivalent to that from the native plasmid (Fig. 3A, lanes 2 and 3). We assayed the stability of this plasmid in B. subtilis over approximately 30 generations of vegetative growth in the absence of antibiotic selection. We assayed in parallel a variant of the plasmid in which we replaced alp7A with an in-frame deletion of the gene, and another variant that contained the pLS20 origin of replication but no alp7AR operon (Fig. 2A).
The plasmid containing both the pLS20 origin of replication and the intact alp7A operon was as stable as pLS20 itself (Meijer et al., 1995), and was retained with no loss at all over the 30 generations of the assay (Fig. 2B). In marked contrast, the plasmid containing the alp7A deletion, mini-pLS20Δ(alp7A), was present in only 55% of the cells at 9.5 generations, and in only 2% of the cells by the end of 32 generations, an 8% loss per generation. The plasmid missing the entire alp7AR operon was also unstable, and was present in only 52% of the cells at the end of 33 generations, a 2% loss per generation (Fig. 2B). These data demonstrated that Alp7A is essential for plasmid stability and that it is very likely a component of a plasmid partitioning system. In many such systems, production of the adaptor DNA-binding protein without its nucleotide-binding partner is more destabilizing than having no partitioning system at all (Łobocka and Yarmolinsky, 1996).
The Alp7A–GFP fusion protein is functional
Actin and the previously characterized bacterial actins are dynamic cytoskeletal proteins. In order to determine whether Alp7A was as well, we examined the behaviour of our C-terminal GFP fusion protein in the context of mini-pLS20. We replaced alp7A on this plasmid with alp7A–gfp (Fig. 2A). Two lines of evidence indicated that the Alp7A–GFP protein was functionally equivalent to Alp7A and was therefore a reliable reporter of its behaviour. First and most importantly, the mini-pLS20alp7A–gfp plasmid was just about as stable as mini-pLS20. After 30 generations of growth in the absence of selection, 97% of the cells still retained the plasmid (Fig. 2C). Immunoblotting revealed that it was the intact Alp7A–GFP fusion protein that was functioning in these cells. The fusion protein was stable; no Alp7A was being generated from proteolytic cleavage (Fig. 3A, lane 5).
Second, the fusion protein complemented mini-pLS20Δ(alp7A) as effectively as Alp7A did in a plasmid stability assay. For this experiment, alp7A and alp7A–gfp were each placed under control of the xylose promoter, each was integrated into the B. subtilis chromosome in single copy via a double recombination event (Fig. 2D) and mini-pLS20Δ(alp7A) was then introduced into each of the two strains. When the transformants were grown in the presence of xylose and assayed after approximately 21 generations, mini-pLS20Δ(alp7A) was found to be present in both strains in about 75% of the cells (Fig. 2E). In the absence of xylose, fewer than 10% of the cells retained the plasmid. Complementation in the PxylA alp7A–gfp strain was again due to Alp7A–GFP itself and not to an Alp7A proteolytic cleavage product; immunoblotting revealed that the fusion protein produced from the chromosome was stable over a range of induction levels (Fig. 2F).
Alp7A is a dynamic cytoskeletal protein
We used fluorescence microscopy to monitor the behaviour of the Alp7A–GFP protein in growing cells of B. subtilis. Nearly all of the cells contained one or more curved filaments (Fig. 3B and C), and in time-lapse experiments these grew and shrank rapidly (Fig. 4A–D and Movie S1). In some cases, a single filament would grow to the length of the cell, then shrink almost to vanishing, and then grow again to its former length. In most cases, the growth or shrinkage was less extensive, but cycles of growth and shrinkage were always present (Fig. 4A–D and Movies S1 and 2). Profiles of several of these filaments from different cells revealed that the rate of growth was a fairly uniform 0.073 ± 0.014 μm s−1, and the rate of shrinkage was 0.14 ± 0.040 μm s−1 (n = 11; Fig. 4A–D and Movie S2). This dynamic instability, a property of eukaryotic microtubules, has also been shown to be a property of the bacterial actin ParM (Garner et al., 2004; 2007; Campbell and Mullins, 2007).
We observed similar filaments and the same dynamic instability when both Alp7A and Alp7A–GFP were produced from the same plasmid, one that we constructed by integration of a plasmid containing alp7A–gfp via a single recombination event into pLS20 itself (Fig. 3A, lane 4; Fig. 3D and E and Movie S3). The filament growth rate was 0.062 ± 0.014 μm s−1, and the shrinkage rate was 0.14 ± 0.061 μm s−1 (n = 8; Fig. 4E–H and Movie S4). The similarity between this profile and that of mini-pLS20alp7A–gfp was consistent with our finding that Alp7A–GFP and Alp7A are functionally equivalent.
Alp7A function requires that it assemble into filaments that are dynamically unstable
Polymerization is critical to actin function. In order to determine whether this was so for Alp7A, we introduced two mutations that, based upon biochemical and structural studies with actin, would be expected to alter the polymerization properties of the protein (Kabsch et al., 1990; Belmont et al., 1999). We focused upon residues whose side chains, as opposed to backbone amides, interact with nucleotide (Kabsch et al., 1990).
Amino acid D212 in Alp7A corresponds to amino acid D154 in actin and is located in the Phosphate 2 sequence (Fig. 1B). The D154 side chain carboxylate interacts with the β and γ phosphates of ATP and with the β phosphate of ADP through a bound divalent cation (Kabsch et al., 1990). A mutation to alanine was introduced into mini-pLS20 and into mini-pLS20alp7A–gfp. The mini-pLS20alp7A(D212A)-gfp plasmid did not give rise to filaments in B. subtilis; instead, the diffuse fluorescence present throughout the entire cell indicated that the mutant protein, although present at the same steady state levels as the wild-type protein (Fig. S2), did not assemble into higher-order structures (Fig. 3F and G). In a plasmid stability assay, the mini-pLS20alp7A(D212A) plasmid was as unstable as mini-pLS20Δ(alp7A) (Fig. 2C). As is the case for actin, its ability to assemble into filaments is essential to the function of Alp7A.
Amino acid E180 in Alp7A corresponds to amino acid Q137 in actin and is located in the Connect 1 sequence (Fig. 1B). The Q137 side-chain amide interacts with the same cation as does the side-chain carboxylate of D154 (Kabsch et al., 1990). A mutation to alanine was introduced into mini-pLS20 and into mini-pLS20alp7A–gfp. The mini-pLS20 alp7A(E180A)-gfp plasmid produced wild-type levels of protein (Fig. S2), but gave rise to filaments in B. subtilis that were unlike those of the wild-type (Fig. 3H–J). Whereas all of the wild-type filaments were contained entirely within a cell, many of the E180A filaments extended from one cell into the next, or even through a row of cells (Fig. 3H). And whereas the wild-type filaments were dynamically unstable, undergoing rapid cycles of polymerization and depolymerization, the E180A filaments were static. In time-lapse experiments, there were no dynamics observed (Movie S5), and in fluorescence recovery after photobleaching (FRAP) experiments, there was no recovery of fluorescence even 1 min after photobleaching (Fig. 3J and Movie S6). In a plasmid stability assay, the mini-pLS20alp7A(E180A) plasmid was nearly as unstable as mini-pLS20Δ(alp7A), and was present in only 18% of the cells at the end of 31 generations (Fig. 2C). As is the case for actin, and also for ParM and AlfA, the ability to assemble into dynamic filaments is essential to the function of Alp7A.
Production of dynamic filaments requires additional elements of pLS20
Our early efforts at intracellular production of Alp7A, in which filaments were observed to form in the absence of any other elements from pLS20 or from the native host, demonstrated that the ability to polymerize into filaments was most likely an intrinsic property of the protein (Fig. 1D). Subsequent experiments supported this conclusion. Alp7A–GFP, so long as it was produced at sufficiently high levels, gave rise to filaments, but to static filaments only (Fig. 5A and Movie S7).
In order to identify any extraneous elements required to produce dynamic filaments, we surveyed the behaviour of Alp7A–GFP in several contexts by time-lapse microscopy. We observed dynamic filaments when Alp7A–GFP was produced in a strain containing pLS20 and this was so whether the alp7A–gfp gene was expressed from the same DNA macromolecule, as in the integrant described above (Fig. 4E–H and Movies S3 and 4) or from the chromosome (data not shown). It was not required that the entirety of pLS20 be present; mini-pLS20 sufficed (Fig. 5B and Movie S8), and mini-pLS20Δ(alp7A) sufficed as well (Fig. 5C and Movie S9). But mini-pLS20Δ(alp7AR) did not; we did not observe dynamic filaments when alp7A–gfp was expressed in a cell containing only mini-pLS20Δ(alp7AR) with no other elements from pLS20 (Fig. 5D). One or more requirements for dynamic Alp7A filaments was therefore contained in 674 bp of pLS20 DNA that was present on mini-pLS20Δ(alp7A) but not on mini-pLS20Δ(alp7AR). Within this 674 bp are alp7R, the second gene of the putative operon, and the 165 bp of DNA that lies directly upstream of the alp7A initiation codon (Fig. 2A).
DNA containing alp7R lowers the critical concentration for Alp7A filament formation
This segment of DNA containing alp7R not only determined whether Alp7A dynamic filaments would assemble, but also at what concentration they formed. We examined the ability of Alp7A–GFP to assemble into filaments at various intracellular concentrations in either the presence or absence of the mini-pLS20Δ(alp7A) plasmid, which has the segment, by counting the number of cells that contained at least one filament. When alp7A–gfp was expressed in the absence of the plasmid, there were no filaments in the cells at xylose induction levels of 0.025% or below; the Alp7A–GFP that was produced accumulated in the cells only as soluble protein (Fig. 6C, D and M). Even at 0.05% xylose, filaments were present in fewer than 5% of the cells (Fig. 6E and M). Only at 0.1% xylose and higher were filaments present in 50% of cells (Fig. 6F, G and M). Yet immunoblotting experiments demonstrated that the steady state levels of Alp7A–GFP increased as expected with increasing concentrations of xylose (Fig. 6A and N). We therefore concluded that there was a critical intracellular concentration that must be attained for Alp7A to polymerize into filaments.
This critical concentration was lowered when mini-pLS20Δ(alp7A) was present in the cell. At 0.05% xylose, nearly 40% of the cells had filaments (Fig. 6J and M). Indeed filaments were present in the cells at xylose concentrations as low as 0.01% (Fig. 6H–L and M). In contrast, in the absence of the plasmid, fewer than 5% of the cells contained filaments at 0.05% xylose, even though physiological levels of Alp7A–GFP were produced (Fig. 6A). For any given concentration of the inducer xylose, the same amount of Alp7A–GFP was produced in both strains (Fig. 6A, B and N). Hence pLS20 DNA containing alp7R and the region upstream of alp7A lowered the critical concentration for Alp7A filament formation.
Alp7A filaments colocalize with plasmids
Because Alp7R is likely to be a DNA-binding protein, it seemed likely that Alp7A filaments assemble on the plasmid. If this were so, each filament could be expected to have a plasmid associated with it in the cell. We tagged the mini-pLS20alp7A–gfp plasmid for fluorescence microscopy by introducing into the plasmid a tandem lac operator array and expressing lacI-cfp from a single copy integrant in the B. subtilis chromosome, and we recorded the relative positions of plasmid foci and Alp7A–GFP filaments in fixed cells. In nearly every case (99%, n = 175), filaments colocalized with plasmid foci (Fig. 7A–F and Movie S10). We also observed complete coincidence of foci and filaments in time-lapse experiments with growing cells (100%, n = 45). Foci were typically found at the ends of filaments as would be expected if filament assembly occurred on the plasmid (Fig. 7A–C, arrowheads), but they could be found in the middle of filaments as well (see below). Further support for the idea that filament formation begins at a plasmid came from tallying the number of plasmid foci and filaments per cell. Although there was little to no correlation between the length of a cell and the number of foci or the number of filaments within it (Fig. 7G and H), there was a relationship between the number of foci and the number of filaments within a cell. As the number of foci per cell increased from 1 up to 10, the number of filaments per cell increased from 1 up to 4 (Fig. 7I). These findings are consistent with a mechanism in which plasmids serve as sites of assembly for Alp7A filaments.
Time-lapse experiments revealed the salient features of the plasmid partitioning mechanism. Separation of plasmid foci was achieved by filament elongation between them, and the rate of separation was consistent with the rate of filament elongation (Fig. 7J and K and Movie S11). But separation was not always a simple binary operation, with a single focus at each end of a filament. For example, a focus at one end of a filament could split, giving rise to two foci that would then be separated from each other by a second elongating filament. This would result in three foci being separated by two growing filaments (not shown). This process generates one focus that appears in fixed cells to be situated in the middle of a single filament (Fig. 7A–F and Movie S10); in reality the focus is bridging two separate filaments.
Alp7A filaments are capable of treadmilling
After plasmids were separated, filaments could remain assembled and fully elongated, but it was not clear if they still retained their dynamic properties. We therefore monitored these filaments after marking them by photobleaching. An example of such an experiment is presented in Fig. 7L. An internal section of the filament was bleached with a laser – the red bracket demarcates the bleached zone – and images were captured over the next 30 s. As polymerization proceeded at the left end of the filament and depolymerization proceeded at the right end, the photobleached zone migrated rightward. Although the position of the filament within the cell was essentially unchanged, addition of new subunits at the left end pushed to the right the subunits already within the filament (Fig. 7L and M). Immediately post bleach, the filament retained its full length as the addition of new subunits at the left end was offset by the loss of subunits from the right end. But by 8 s post bleach, depolymerization had outpaced polymerization and the process of filament disassembly was underway (Fig. 7L, line c). Figure 7N illustrates the same behaviour in a cell containing CFP-LacI tagged plasmids at the filaments ends. Here photobleaching of the filament also resulted in the bleaching of part of the cytoplasmic Alp7A–GFP pool, so Alp7A filament polarity could be inferred from the observation that new (and distinctly dimmer) subunits were incorporated only at the right end of the filament. As in the filament of Fig. 7L, fluxing occurred as the bleached subunits (red bracket) were pushed to the left by the addition of subunits (white bracket) to the right end. The data of Fig. 7L and N indicate that in addition to undergoing periods of rapid growth and shrinkage that are characteristic of dynamic instability, Alp7A–GFP filaments can also treadmill. We observed treadmilling only in fully elongated filaments. Plasmid foci were present at the ends of these filaments, suggesting that treadmilling occurs after plasmid separation.
Actin-like proteins are widespread in bacteria and extremely divergent
A phylogenetic analysis led us to identify more than 35 distinct families of actin-like proteins (Alps) in bacteria. Some of these families are so divergent in sequence from known actins that their connection to the actin superfamily had until now gone unnoticed. The functions of most of the Alps have yet to be determined, but their genes are found on phage, plasmids, integrating conjugative elements, and in a few cases, on bacterial chromosomes. The Alps are therefore likely to participate in a variety of processes. Alp7A, which we investigated in depth, is a plasmid partitioning protein, as are the previously characterized ParM and AlfA. But not all of the Alps are involved in plasmid partitioning. Alp8A is encoded on a plasmid but is not required for its stability (data not shown). Other Alp8 family members are encoded on integrative conjugative elements that do not replicate autonomously, and so would not require a partitioning machinery. Members of the three other previously characterized Alp families function in cell shape determination (MreB), in cell division (FtsA) and in organelle positioning (MamK).
Despite the very tenuous connection of these sequences to eukaryotic actin, we were able to determine that a member of one of the most divergent of these families is indeed a bacterial actin. Alp7A formed filaments within the cell and these filaments exhibited two dynamic behaviours, dynamic instability and treadmilling. Treadmillling is a behaviour associated with eukaryotic actin and has also been reported in the C. crescentus MreB (Kim et al., 2006). Dynamic instability is a fundamental property of the bacterial actin ParM (Garner et al., 2004).
Because our Alp7A–GFP fusion protein retained its function and could be used interchangeably with Alp7A itself, we could easily correlate Alp7A function with its behaviour in the cell and assess whether the actin properties of Alp7A were required for its function. We found that mutations in two amino acids that would be predicted to interact with nucleotide on the basis of actin biochemistry and structural biology disrupted Alp7A polymerization dynamics. The D212A mutation, which abolished filament formation, was indistinguishable from a null mutation in a plasmid stability assay. The E180A mutation, which permitted filaments to form but eliminated their dynamic properties, was almost as crippling. The analogous mutation in ParM (E148A) eliminates ATP hydrolysis, and leads to stable filament formation (Garner et al., 2004). It follows that in order to function as a plasmid partitioning protein, Alp7A must behave as a bacterial actin: it must polymerize into filaments with dynamic properties.
Plasmid partitioning by dynamic Alp7A filaments
ParM filaments form spontaneously from the purified ParM protein in the presence of ATP, and these filaments are dynamically unstable; no auxiliary or nucleating factors are required (Garner et al., 2004). Addition of the ParR/parC complex, which corresponds to Alp7R and the upstream sequence in the Alp7A system, regulates ParM assembly by suppressing dynamic instability and stabilizing the filaments (Garner et al., 2007). ParM filaments are typically not observed in vivo in the absence of the ParR/parC complex because they are so unstable (Campbell and Mullins, 2007; Møller-Jensen et al., 2002). When Alp7A was produced at its physiological level from an inducible promoter with no other elements of plasmid pLS20 present in the cell, we detected no filaments. We detected filaments at the physiological concentration of Alp7A only if there was also present in the cell a small segment of pLS20 DNA that contains both the alp7R gene and the 165 bp directly upstream of the alp7A initiation codon. Alp7R-bound plasmid might function to stabilize transient Alp7A filaments that form spontaneously and that would otherwise quickly fall apart and therefore be undetectable, as has been demonstrated for ParM. Alternatively, Alp7R-bound plasmid might serve as a nucleation factor that is required to initiate Alp7A filament formation at its physiological concentration. By either mechanism, every Alp7A filament would be expected to be associated with a plasmid, and this is in fact what we observed.
The colocalization experiments of Fig. 7 enable us to highlight some of the basic features of the Alp7A plasmid partitioning mechanism, from which we present a model. The filaments, once initiated at the plasmid, are dynamically unstable, and repeatedly grow and shrink. Elongation in conjunction with dynamic instability enables the filament to search again and again for a second plasmid. Should the free end of a filament encounter a second plasmid, continued elongation drives the two plasmids apart. It is possible that capture of the free end of an Alp7A filament by a second plasmid abolishes dynamic instability, as is the case for ParM. When the plasmids are as far apart from each other as the cell boundaries permit, and the filaments can elongate no longer, they treadmill. We propose that treadmilling serves as an idle to keep plasmids apart after they have been separated from each other. When the fully elongated filaments disassemble, the process can begin again.
This partitioning mechanism resembles fundamentally the ParM partitioning mechanism: plasmids are associated with dynamic filaments, and it is the elongation of the filaments between the plasmids that brings about their separation (Gerdes et al., 2004; Møller-Jensen et al., 2002; 2003; Campbell and Mullins, 2007; Garner et al., 2007; 2004; Salje et al., 2009). Nevertheless there may turn out to be differences between Alp7A and ParM, in particular with respect to the dynamic behaviour of the filaments. Both Alp7A and ParM show dynamic instability, but we found that Alp7A filaments can also treadmill; there has to date been no report of treadmilling of ParM. Alp7A also differs from AlfA, another plasmid partitioning protein from B. subtilis. Unlike Alp7A and ParM, AlfA does not show dynamic instability in vivo (Becker et al., 2006). It appears then, that even Alps that perform identical functions in the cell may differ in some of their properties. This might seem counterintuitive, but is actually unsurprising given that these Alps are so distantly related to one another.
Are there many more actins out there?
What may be said for the many other families of proteins that emerged from our phylogenetic analysis? Are all of these Alps actins as well? We chose three, more or less at random, based upon availability of DNA and tractability of the host bacterium, and all three formed filaments in E. coli. And the one that we investigated in depth, Alp7A, is almost certainly an actin, although verification of this will require a crystal structure. If all of these other Alps do turn out to be actins, the number of known bacterial actin families will be increased nearly eightfold, and an already enormous superfamily that includes in addition to actin, the hexokinase-like sugar kinases and the Hsp70 proteins will be that much expanded (Hurley, 1996; Bork et al., 1992). There are many more Alps to be uncovered in this manner, and the number will continue to increase as more sequence information becomes available.
The Alps fall into more than 40 highly divergent families; each family shares less than 30% amino acid identity with each of the others. Among the three Alps that we worked with, the amino acid identity ranges from 15% to 21% with respect to the others, and is uniformly 12% with respect to actin. It is therefore unsurprising that many of these proteins have not been recognized as actins to date. For MreB, the quintessential bacterial actin, coming to this recognition, took many years (Egelman, 2001). Like Alp6A, Alp7A and Alp8A, MreB shares only 15% identity with actin (van den Ent et al., 2001).
Eukaryotic actin is extremely well conserved. It has been proposed that the eukaryotic sequence is evolutionarily constrained by the very many proteins with which actin must interact or by the intricate structural requirements of one or more substrates that actin adopts in the context of the filament (Galkin et al., 2002). While the eukaryotic actins have been confined to a very tight evolutionary space, the opposite appears to be the case for the Alps. Their great divergence attests to a long evolutionary history or to a rapidly paced evolution, and very likely to both. The actin lineage is ancient, and probably goes back to the early history of life on earth (Doolittle and York, 2002). Most of the Alps are found on mobile genetic elements such as plasmids and phages (Fig. 1A and Table S1A). Mobile genes such as these typically evolve faster than the genes of their host organisms (Drake et al., 1998). We do not know what most of these genes do, but their products would not be expected to play the central role that actin plays in eukaryotic cells, and so similar evolutionary constraints would not apply.
The actins then are the products of a broad spectrum of evolutionary selection. At one end of the spectrum is eukaryotic actin, the uniform and unvarying outcome of the enormous selective pressure imposed by its critical function in the cell. At the other end are the Alps, the expansive and diverse outcome of the fewer and simpler demands on their functions. Even among the Alps though, the signature property of actin would appear to have been conserved: all of the Alps that have been characterized to date have been shown to polymerize. But there is variety in filament structure, in the kinetics of filament formation and in cellular function. Studies of other Alp family members should give us a still greater appreciation of how remarkably adaptable the actin prototype has proven to be in bacteria.
The AlfA sequence was used to begin the blast iterations series; the same sequences were retrieved if other bacterial actins were used, although not necessarily in the same order.
Sequences were aligned with TCoffee and ClustalW, and phylogenetic trees were constructed with ClustalW. Figure 1A is a bootstrap consensus tree of 1000 trees that was generated by the neighbor-joining clustering method. Similar trees were obtained regardless of the method used. The cut-off for assignment to a family was 30% sequence identity.
Standard techniques of molecular biology were used. Genomic DNA was purified from Bacillus with a modification of a protocol developed for Gram-negative bacteria (Neumann et al., 1992). Other nucleic acid purifications were done with commercial kits manufactured by Qiagen or Invitrogen. Oligonucleotide primers were synthesized by Allele Biotechnology and Pharmaceuticals or by Integrated DNA Technologies. PfuUltra High-Fidelity Polymerase, which was used for nearly all PCR amplifications, was obtained from Stratagene. Amplifications were carried out in a Mastercycler EP (Eppendorf). Restriction endonucleases were obtained from New England Biolabs unless otherwise noted. Shrimp alkaline phosphatase was obtained from Roche Diagnostics GmbH, and T4 DNA ligase from New England Biolabs, RNAase was obtained from Qiagen and DNAase from Invitrogen. Other biochemicals and chemicals were obtained from Fisher, VWR or Sigma. Plasmids were introduced into E. coli strains DH5α, MG1655 or TOP10 by electroporation with a Gene Pulser Xcell (Bio-Rad) or by transformation of chemically competent cells (Hanahan, 1985). DNA sequencing was performed by Eton Bioscience or by Genewiz. Primer sequences are provided in Table S2 of the Supporting information.
Sequencing of the alp7AR operon
Semidegenerate PCR was used to amplify the latter part of the alp7A gene and the remainder of the alp7AR operon (Jacobs et al., 2003). Amplicons were cloned into the pCR2.1-TOPO vector (Invitrogen) and submitted for sequencing. These sequence data have been submitted to the DDBJ/eMBL/GenBank databases under the accession number GQ337014. The sequences of Alp7A and Alp7R are presented in Fig. S1.
Plasmids and plasmid constructions
Alp6A. Bacillus thuringiensis phage 0305φ8-36 DNA was obtained from Stephen Hardies and Julie Thomas at the University of Texas Health Science Center, San Antonio, Texas.
Plasmid pPAU12 (pPxyl alp6A-gfp) was constructed from plasmid pPAU11, which contains a fusion of gfp to alp6A. pPAU11 was constructed by PCR amplification of B. thuringiensis phage 0305φ8-36 DNA (Thomas et al., 2007) with oligonucleotide primers P1 and P2, restriction of the amplicon with KpnI and ClaI, and ligation of the product to plasmid pMUTIN-GFP+ (Kaltwasser et al., 2002) restricted with KpnI and ClaI. The cloned segment includes 41 bp upstream of the alp6A initiation codon. pPAU11 DNA was amplified with oligonucleotide primers P3 and P4, the amplicon was restricted with KpnI and ligated to pWH1520 (Rygus and Hillen, 1991) restricted with KpnI.
Alp7A. Plasmid pAID3107 (pPxyl alp7A–gfp) was constructed from plasmid pAID3068, which contains a fusion of gfp to alp7A. pAID3068 was constructed by PCR amplification of genomic DNA from strain IFO3335 with oligonucleotide primers P7 and P8, restriction of the amplicon with KpnI and ClaI, and ligation of the product to plasmid pMUTIN-GFP+ restricted with KpnI and ClaI. The cloned segment includes 731 bp upstream of the alp7A initiation codon. pAID3068 DNA was amplified with oligonucleotide primers P9 and P10, the amplicon was restricted with KpnI and SphI, and ligated to pWH1520 restricted with KpnI and SphI, to produce pAID3107.
Plasmid pAID3129 (mini-pLS20) was constructed by PCR amplification of genomic DNA from B. subtilis natto strain IFO3335 with oligonucleotide primers P11 and P12, restriction of the amplicon with NsiI and NheI, and ligation of the product to plasmid pHW1520 restricted with NsiI and NheI. The 3501 bp cloned segment contains a fragment of orfA, prematurely terminated at amino acid 141, the pLS20 origin of replication, and the orfBC (alp7AR) operon through its transcription terminator. Plasmid pAID3147 [mini-pLS20Δ(alp7A)] was constructed via a modification of the standard PCR-based site-directed mutagenesis protocol with pAID3129 as template and mutagenic oligonucleotide primers P13 and P14 (Wang and Malcolm, 1999). In pAID3147, alp7A is replaced by an in-frame deletion that consists of an AvrII site flanked by the first four and last five codons of the gene. Plasmid pAID3171 [mini-pLS20Δ(alp7AR)] was constructed by restriction of pAID3129 with NheI, fill-in of the 5′ overhang with T4 DNA polymerase, partial digestion with SmaI and monomolecular ligation of the 8387 bp fragment. pAID3171 contains the prematurely terminated orfA fragment, the origin of replication and pLS20 sequences through 166 bp upstream of the alp7A initiation codon.
Plasmid pEB416 [mini-pLS20 (lacO)x] was constructed by introducing into pAID3129 a fragment containing a spectinomycin-resistance gene flanked by lacO arrays. This fragment was constructed by modifying plasmid pLAU43 (Lau et al., 2003), which contains arrays of 120 lacO operators on either side of a gene that codes for kanamycin resistance. Plasmid pSE380 (Invitrogen) was restricted with SalI and XbaI, and the 118 bp fragment derived from the multiple cloning site was ligated to pLAU43 restricted with SalI and XbaI. The kanamycin-resistance gene in the resulting plasmid, pRL153, was then replaced with one for spectinomycin resistance from plasmid pMDS13 (Sharp and Pogliano, 2002) by amplification of pMDS13 with primers P15 and P16, restriction of the amplicon with NsiI, and ligation of the product to pRL153 restricted with NsiI. Restriction of the resulting plasmid with BamHI generated the fragment that was ligated to pAID3129 restricted with BglII.
Plasmid pAID3205 (pPxyl alp7A) was constructed from pAID3107. pAID3107 was restricted with EcoRI in the presence of ethidium bromide, then with EagI, and the two 5′ overhangs were filled in with T4 DNA polymerase. Monomolecular ligation of the resulting 9218 bp fragment produced a template for site-directed mutagenesis with oligonucleotide primers P17 and P18, which modified the blunt end junction to match the transcription termination sequences to that of alp7A–gfp in pAID3107.
Plasmid pAID3195 (mini-pLS20alp7A–gfp) was constructed by ligating the 7706 bp BspEI-MluI restriction fragment from pAID3147, the 2631 bp BspEI-SpeI restriction fragment from methylated pAID3068, and the SpeI-MluI restricted amplicon generated by PCR amplification of pAID3147 with oligonucleotide primers P19 and P20. In pAID3147, the Δalp7A in-frame deletion and alp7AR intergenic region is interposed between alp7A–gfp and alp7R in order to place alp7R into its native translational context.
The alp7AR mutations D212A and E180A were constructed via standard PCR-based site-directed mutagenesis (Papworth et al., 1996) with template pAID3205 (for D212A) or a smaller variant of pAID3129 (for E180A) with oligonucleotide primers P21 and P22 (D212A) and oligonucleotide primers P23 and P24 (E180A). The mutations were then introduced into pAID3129 and pAID3107 by swapping in a 695 bp AgeI restriction fragment.
Plasmid pAID3118 (pPT7 His6-alp7A) was constructed by PCR amplification of genomic DNA from strain IFO3335 with oligonucleotide primers P25 and P26, cloning of the amplicon into the pCR-Blunt II-TOPO vector (Invitrogen), restriction of the resulting plasmid with NheI and ligation of the 1179 bp fragment to plasmid pET-28a(+) (Novagen) restricted with NheI.
Alp8A. Plasmid pEB400 (pPtrc[Rts1 orf250]-gfp) was constructed by PCR amplification of genomic DNA from E. coli strain ER1648 with oligonucleotide primers P27 and P28, restriction of the amplicon with KpnI and PstI, and ligation of the product to pDSW210 (Weiss et al., 1999) restricted with KpnI and PstI. The promoter in pDSW210 is a variant of the Ptrc promoter.
Bacterial strains and strain constructions
Bacillus subtilis natto strain IFO3335 (BGSC 27E1) (Tanaka and Koshikawa, 1977) was obtained from the Bacillus Genetic Stock Center at The Ohio State University, Columbus, OH. E. coli strain ER1648 containing plasmid Rts1 (Murata et al., 2002) was obtained from Tetsuya Hayashi at the University of Miyazaki, Miyazaki, Japan. B. subtilis strains BEST2125 and BEST40401 (Itaya et al., 2006) were obtained from Mitsuhiro Itaya at the Mitsubishi Kagaku Institute of Life Sciences, Tokyo, Japan.
All physiology and microscopy experiments were carried out at 30°C in B. subtilis strain PY79 (Youngman et al., 1984) or in E. coli strains DH5α, MG1655, or TOP10 (Invitrogen). Strain JP3100 (pLS20cat/PY79) was constructed by first conjugating plasmid pLS20cat from strain BEST40401 into strain BEST2125, and from the resulting exconjugant into PY79 (Itaya et al., 2006). Strain JP3104 (JP3100 pLS20cat alp7A::pAID3068) is an integrant of plasmid pAID3068 into the pLS20cat plasmid resident in JP3100. Strain JP3161 (PY79 thrC::xylR+PxylA alp7A–gfp) was constructed by integration into the PY79 chromosome of a segment of plasmid pAID3107 containing the xylR gene and PxylA alp7A–gfp. A 3918 bp segment was amplified from pAID3107 with primers P29 and P30, the amplicon was restricted with BglII, and the product was ligated to B. subtilis chromosomal integration vector pDG1664 (Guérout-Fleury et al., 1996) restricted with BamHI, to match the transcriptional orientation of the threonine operon on the vector. The cloned segment was then integrated into the PY79 chromosome at thrC by a double recombination event. The same strategy was used to construct strain JP3206 in which a 3180 bp segment of plasmid pAID3205 containing the xylR gene and Pxyl alp7A is integrated into the PY79 chromosome.
Strain EBS1340 (PY79 amyE::PxylA[lacI–cfp3A]) was constructed by integrating into the PY79 chromosome a segment from plasmid pEB387, a derivative of the B. subtilis chromosomal integration vector pDG1662 (Guérout-Fleury et al., 1996). pEB387 was constructed from plasmid pMDS78, a derivative of pDG1662 that contains PspoIIR gfp, the gfp gene under control of the B. subtilis spoIIR promoter (Sharp and Pogliano, 2002). The spoIIR promoter region in pMDS78 was replaced with the spoIIE promoter region by PCR amplification of the spoIIE promoter region from PY79 with primers P31 and P32, restriction of the amplicon with BamHI and EcoRI, and ligation of the product to pMDS78 restricted with BamHI and EcoRI. The gfp gene in this intermediate plasmid was then replaced with the cfp3A gene by PCR amplification of the gene from pSCFP3A-C1 (Kremers et al., 2006) with primers P33 and P34, restriction of the amplicon with SpeI and EagI, and ligation of the product to the intermediate plasmid restricted with SpeI and EagI. The lacI fusion to cfp3A was constructed in this second intermediate plasmid. The lacI gene lacking the coding sequence for the last 11 amino acids was amplified from pMUTIN-GFP with primers P35 and P36, the amplicon was restricted with SpeI and BamHI, and the product was ligated to the second intermediate plasmid restricted with SpeI and BamHI. The spoIIE promoter in this plasmid, pEB307, was then replaced with Pxyl by PCR amplification of plasmid pEA18 (Quisel et al., 1999) with primers P37 and P38, restriction of the amplicon with BglII and EcoRI, and ligation of the product to pEB307 restricted with BglII and EcoRI. Lastly, the ribosome-binding site for the lacI–cfp3A fusion in this plasmid, pEB384, was replaced with an optimized version generated by amplification of the fusion from pEB384 with primers P34 and P39, digestion of the amplicon with HindIII, and ligation to pEB384 digested with HindIII. P39 introduces the modified ribosome-binding site and also appends eight codons (MKNIEKVS) to the beginning of the lacI gene. The Pxyl lacI–cfp3A gene fusion was then integrated onto the PY79 chromosome at amyE by a double recombination event, to produce strain EBS1340.
All other B. subtilis strains were constructed by standard transformation of PY79 or derivatives of PY79 with the plasmids described (Dubnau and Davidoff-Abelson, 1971). pLS20cat was introduced into strains by conjugation.
Media for strains containing pLS20cat was supplemented with 5 μg ml−1 chloramphenicol. Media for strains containing derivatives of pWH1520 were supplemented with 100 μg ml−1 ampicillin or carbenicllin for E. coli, or with 10 μg ml−1 tetracycline for Bacillus. Erythromycin was used at 2 μg ml−1 for Bacillus, kanamycin at 50 μg ml−1 for E. coli and spectinomycin was used at 100 μg ml−1 for either Bacillus or E. coli.
Plasmid stability and plasmid stability complementation assays
Shake flask cultures in Luria–Bertani (LB) medium were inoculated from small starter cultures in LB medium supplemented with 5 μg ml−1 chloramphenicol or 10 μg ml−1 tetracycline. Cultures were aerated at 250 r.p.m. and maintained in exponential growth at 30°C by iterative 1/60 dilution into flasks containing prewarmed medium at early exponential phase (OD600 = 0.1 or 0.2), corresponding to approximately six generations. Growth was taken to the end of 30 generations. At each dilution, samples were plated on non-selective medium, and 100 colonies were tested for retention of antibiotic resistance. Generation times were calculated from each interval and the mode value was applied to the entire growth course.
For complementation assays, starter and experimental cultures contained an appropriate amount of xylose or glucose, growth was continued for approximately 20 generations, and platings were done only at t0 and at the end of the experiment.
Hexahistidine-tagged Alp7A was recovered from strain JP3118 as inclusion bodies after a 3 h induction at 30°C. The cells were lysed as described (Derman et al., 1993), treated with DNase I (Invitrogen) and the postlysis pellets containing the inclusion bodies were washed twice with water and then twice with a buffer consisting of 300 mM NaCl, 12.5 mM imidazole, 50 mM NaxHyPO4, pH 8.0. The washed pellets were dissolved in the same buffer containing 8 M urea, the solution was centrifuged at 20 000 g for 30 min, and the denatured Alp7A was purified from the supernatant by nickel affinity chromatography as described except that 8 M urea was present throughout (Lim et al., 2005). Fractions containing Alp7A were dialysed against PBS and the dialysed protein was used for antibody preparation. Polyclonal antibodies were generated in rabbits by Antibodies Incorporated.
Proteins were electrotransferred from polyacrylamide gels to PVDF membranes, and probed with the polyclonal antiserum raised against Alp7A and an anti-rabbit IgG linked to HRP (GE Healthcare). Immunoblots were developed with the ECL Plus Western Blotting Detection System (GE Healthcare), visualized with a Typhoon 9400 Variable Mode Imager (GE Healthcare) and quantified with ImageQuant Software, version 5.0 (GE Healthcare).
Fixed cells or cells from late exponential cultures were pelleted, resuspended in roughly 10% of the original volume of supernatant, affixed to a poly L-lysine-coated coverslip and visualized with a DeltaVision Spectris Restoration Microscopy System (Applied Precision) with an Olympus IX70 Inverted System Microscope and a Photometrics CoolSNAP HQ CCD camera. Data were collected and analysed with DeltaVision SoftWoRx Image Analysis Software. Seven or eight images were collected as a stack of 0.15 μm increments in the z-axis. Images were deconvolved for 10 cycles in enhanced ratio mode. Deconvolved images are presented unless otherwise indicated.
For time-lapse imaging, growing cells were inoculated directly from a fresh colony onto a 1.2% agar or agarose pad containing 20% or 25% LB medium and appropriate antibiotics and inducers. The slide was incubated at 30°C and imaged without sectioning at uniform intervals, typically 1, 3 or 5 s, in the Weather Station temperature-controlled chamber outfitted to the microscope (Precision Control). Images were deconvolved as above. The SoftWoRx Image Analysis Software was used to measure filament lengths.
For photokinetics experiments (FRAP), a 0.5 s pulse at 50% power was delivered from the Quantifiable Laser Module (488 nm) outfitted to the microscope (Applied Precision), and the field was then imaged at uniform intervals as for time-lapse. Three images were taken prior to bleaching. Images were deconvolved as above.
FM 4-64 (Molecular Probes/Invitrogen) was present in slide preparations at 2 μg ml−1 and in agar pads at 0.2 μg ml−1 (Pogliano et al., 1999).
Co-ordinated Alp7A microscopy and protein quantification
For each strain, a fresh single colony was dispersed in 1 ml LB medium, 100 μl of the suspension was used to inoculate one or more 6 ml cultures of LB medium containing any selective antibiotics, and the cultures were rolled at 30°C. In early exponential phase, the cultures were induced with an appropriate amount of xylose. At the end of 1 h, at which time the culture had typically attained an OD600 of between 0.4 and 0.5, 0.5 ml of the culture was added to 20 μl of 1 M NaxHyPO4 pH 7.4, and the cells were then fixed at room temperature for 20 min with 0.0063% glutaraldehyde in 2.7% paraformaldehyde. The fixed cells were washed three times with PBS, resuspended in PBS and examined by fluorescence microscopy.
At the same time, 1 ml of the culture was added to 1 μl of a protease inhibitor cocktail (Sigma P2714, reconstituted according to the manufacturer's instructions), and PMSF was added to 150 μg ml−1. The cells were pelleted, frozen in a dry ice/ethanol bath and stored overnight at −70°C. The thawed cells were resuspended in 60 μl of a buffer consisting of 40% sucrose, 1 mM EDTA, 33 mM TrisCl pH 8.0 with protease inhibitors as above, and treated with 1 mg ml−1 lysozyme at 37°C for 10 min. An equal volume of SDS-PAGE sample preparation buffer with 5% β-mercaptoethanol was added to the lysate, and the samples were heated at 80°C for 10 min. Proteins were fractionated on SDS-PAGE and immunoblotted.
Celia Ebrahimi constructed plasmid pAID3118 and carried out preliminary expression trials. Tiffany Dunbar collected the images of Fig. 7K. Rachel Larsen constructed plasmid pRL153. We thank Daniel Ziegler, Mitsuhiro Itaya, Stephen Hardies, Julie Thomas and Tetsuya Hayashi for strains, bacteriophage DNA and plasmids. We are especially grateful to Daniel Ziegler at the Bacillus Genetic Stock Center for speedily tracking down the correct version of strain IFO3335. We thank Arshad Desai and Dyche Mullins for many helpful discussions. This material is based on work supported under grants to J.P. from the NIH (R01-GM073898).